svakulenk0/ODExploration_data

Open Data Exploration Dataset

Summary

26 chat-based student conversation transcripts about the content of the two Austrian open data portals:

• data.gv.at (13 conversations)
• opendataportal.at (13 conversations)

16 conversations included elements of exploration with listing of available categories, dataset samples, facets, etc., while other 10 were purely search-based interactions focused on addressing specific questions directly. 24 conversation transcripts as samples with a successful interacion outcome, i.e. resulting in satisfied information need and positive user feedback, 2 - unsuccessful interacion outcome.

Annotations

The transcripts are manually annotated with the dialog act types (intents/responses) and the speaker identifier (agent/User). The new lines separators were inserted within the same message to indicate the spans of text with different message types.

Annotation template: message_type>speaker_turn>message e.g. greeting>E>Hi?

We differentiate between 15 different dialog act types, which correspond to the basic operations available for data exploration.

1. Dialog acts used by both Seeker and Intermediary:

    * greeting(), indicates common start of the conversation, e.g. "hello"
    * confirm(), explicitly confirms the direction for exploration
    * verify(), prompts to confirm the direction for exploration or the results
    * success(), indicates end of the conversation with a successful outcome
    * prompt(link), suggests/asks for a direct link to the dataset
   

   Dialog acts of an Intermediary:

    * prompt(keywords), indicates a general information request
    * list(keywords), indicates available options on different levels (attributes, items)
    * bool(data), reports existance of the requested set of items (datasets/columns/attributes)
    * top(keywords), reports the top (most frequent) items (rows/attributes)
    * count(data), reports the count of items (rows/datasets/attributes)
    * link(dataset), reports the direct link to a dataset
   

   Dialog acts of a Seeker:

    * question(data), indicates a general information request, e.g. "what data do you have?"
    * set(keywords), indicates the choice of a specific direction for exploration, e.g. "have you got smth on population statistics?"
    * reject(), explicitly rejects the direction for exploration
    * more(), indicates a request for more items from the same equivalence class
   

2. Dialog turn separators:

• A for the agent utterance
• U for the user utterance

3. [[]] indicates concept span

Codes for span annotations:

• [[]]H* - greeting;
• [[]]G* - question about the data availability/need;
• [[]]F* - attribute (facet) as a selection option for the next exploration direction, e.g. category;
• [[]]E* - entity option (facet value) for exploration, e.g. finance;
• [[]]Q* - cardinality of the corresponding item set;
• [[]]R* - unique identifier of an item, e.g. title or link to the dataset;
• [[]]+* - positive feedback;
• [[]]-* - negative feedback.

Average number of concepts per utterance 2

Maximum number of concepts per utterance 16

Process models

Produced with ProM: Inductive Visual Miner and Declarative model?

License

All work in this repository is under the MIT license

Acknowledgement

If you find this work useful, feel free to cite us:

@article{DBLP:journals/corr/abs-2012-03704,
  author       = {Svitlana Vakulenko and
                  Vadim Savenkov and
                  Maarten de Rijke},
  title        = {Conversational Browsing},
  journal      = {CoRR},
  volume       = {abs/2012.03704},
  year         = {2020},
  url          = {https://arxiv.org/abs/2012.03704},
}